Related Experiment Video
Updated: May 22, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.7K
A generalised vision transformer-based self-supervised model for diagnosing and grading prostate cancer using
Abadh K Chaurasia1,2, Helen C Harris3, Patrick W Toohey4
1Menzies Institute for Medical Research, University of Tasmania, Hobart, TAS, Australia. awadhaiims@gmail.com.
Prostate Cancer and Prostatic Diseases
|March 15, 2025
Summary
An AI system accurately diagnoses and grades prostate cancer from histological images, overcoming pathologist variability. This digital pathology tool shows strong performance in external validation for clinical use.
Area of Science:
- Digital Pathology
- Artificial Intelligence in Oncology
- Histopathology Image Analysis
Background:
- Gleason grading is crucial for prostate cancer prognosis but suffers from subjective variability.
- This variability can impact clinical decision-making and patient outcomes.
- Developing objective diagnostic tools is essential for improving prostate cancer management.
Purpose of the Study:
- To develop and validate a generalized AI system for prostate cancer diagnosis and grading.
- To assess the AI system's performance on diverse datasets including tissue microarray cores and whole slide images.
- To evaluate the AI system's ability to distinguish benign from malignant tissue and classify Gleason patterns.
Main Methods:
- Analysis of eight prostate cancer datasets (12,711 images, 3648 patients) using tissue microarray cores and whole slide images.
- Application of the Macenko method for color normalization to ensure image consistency.
- Training of multi-resolution binary and multi-class classifiers for tissue classification and Gleason pattern sub-categorization.
- External validation on 11,132 images from 2176 patients to determine International Society of Urological Pathology (ISUP) grade.
Main Results:
- The AI system achieved high performance in distinguishing benign from malignant tissue (κ=0.967 internally, 0.876-0.995 externally).
- The multi-class classifier accurately distinguished Gleason patterns (GP3, GP4, GP5) with an overall κ=0.841.
- Performance on an external dataset compared favorably to an independent pathologist (κ=0.752 for four classes).
Conclusions:
- A self-supervised Vision Transformer (ViT)-based AI model effectively diagnoses and grades prostate cancer from histological images.
- The AI system demonstrates robustness and clinical applicability in digital pathology settings.
- This technology offers a potential solution to subjectivity in prostate cancer grading and aids clinical decision-making.
